STIGMA: Single-cell tissue-specific gene prioritization using machine learning.

Area of Science:

• Genomics
• Developmental Biology
• Computational Biology

Background:

• Clinical exome and genome sequencing advance disease genetics, but many genes remain uncharacterized, hindering variant interpretation.
• Existing gene prioritization methods do not account for cell-specific gene expression heterogeneity within tissues.

Purpose of the Study:

• To introduce STIGMA (single-cell tissue-specific gene prioritization using machine learning), a novel framework for prioritizing candidate genes in rare congenital diseases.
• To leverage single-cell RNA sequencing (scRNA-seq) data to capture cell-type-specific gene expression dynamics during organogenesis.

Main Methods:

• Applied STIGMA to mouse limb and human fetal heart scRNA-seq datasets.
• Trained machine learning models to learn temporal gene expression patterns across cell types.
• Prioritized candidate genes and variants associated with congenital malformations.

Main Results:

• STIGMA identified 469 variants in 345 genes for congenital limb malformations, highlighting UBA2.
• For congenital heart defects, 34 genes with nonsynonymous de novo variants (nsDNVs) were detected in 7,958 individuals, including the Prdm1 ortholog.
• Demonstrated STIGMA's ability to prioritize tissue-specific genes by analyzing cell population heterogeneity.

Conclusions:

• STIGMA effectively prioritizes candidate genes for rare congenital diseases by integrating scRNA-seq data.
• The framework's capacity to model cell-specific expression heterogeneity enhances the discovery of disease-associated genes and causal variants.